8000 GitHub - GeminiLight/AgenticConsensus: A shared, versioned project state for human–AI coding collaboration. Connect intent, code, evidence, and authorized human decisions. · GitHub
[go: up one dir, main page]

Skip to content

About

A shared, versioned project state for human–AI coding collaboration. Connect intent, code, evidence, and authorized human decisions.

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

AgenticConsensus

Paper · Website · Personal website

AgenticConsensus is a shared, versioned project state for human–AI coding collaboration.

AgenticConsensus connects a change request, its interpretation, code changes, requirements, evidence, and an authorized human decision. It helps a reviewer inspect what changed, which requirements are covered, and what still needs checking.

Getting started · Architecture

What it does

  • Reviews immutable Git revisions or a read-only snapshot of a GitHub pull request.
  • Tracks requirements across revisions and keeps evidence bound to the version it checks.
  • Presents code, evidence coverage, open risks, and a connected audit graph.
  • Records authorized human decisions with version checks and an audit trail.
  • Offers optional evidence-grounded model advice. Model output cannot approve a change.

This is an early, local-first research prototype. Source analysis currently focuses on Python imports; automated realization supports a bounded Python import substitution. A recorded decision does not merge a GitHub pull request.

Run locally

Requires Python 3.11+ and Git. No model key is needed for the core demo.

git clone https://github.com/GeminiLight/AgenticConsensus.git
cd AgenticConsensus
python3 -m venv .venv
.venv/bin/python -m pip install -e '.[dev]'
.venv/bin/agentic-consensus demo --reset
.venv/bin/agentic-consensus serve

Open http://127.0.0.1:8765 and create the first administrator account. For the seeded demo, use security-lead@example.test as the audit identity so it matches the synthetic project's authority grants. To reopen the workspace, run serve again; demo --reset rebuilds the demo data.

The demo illustrates a change whose tests pass while a governed architectural relation remains unverified. Inspect its evidence and blockers before recording a decision. Optional model configuration is documented in .env.example.

Runtime data stays in the local var/ directory. Use the default localhost address; see SECURITY.md for the deployment boundary.

Paper and citation

Scaling Human-AI Coding Collaboration Requires a Governable Consensus Layer is an arXiv preprint, first submitted on 20 April 2026. The website describes the newer research and implementation; the linked preprint is an earlier version. Use CITATION.cff or the website's BibTeX citation to cite the paper.

Development and website

.venv/bin/python -m pytest
node --test tests/*.test.cjs
.venv/bin/python -m build

The official website lives in website/ in this repository. Changes to that directory on main publish through GitHub Pages. See website/README.md for preview and maintenance instructions.

See CONTRIBUTING.md for contributions and docs/usage.md for the full workflow.

Licensed under Apache-2.0. See LICENSE and THIRD_PARTY_NOTICES.md.

About

A shared, versioned project state for human–AI coding collaboration. Connect intent, code, evidence, and authorized human decisions.

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages

0